They’re not really separate advantages, though. They describe the same, single advantage.

I’ve spent my career watching the attention economy emerge from the inside, building strategy around where an audience actually comes from. And I’ve watched this ground shift twice. Facebook gradually cut off traffic to publishers and websites after 2016. Then a lot of newsrooms laid people off chasing video numbers that Facebook later admitted were inflated.

Subscribe now

Now Google is pulling back search referrals, with its new AI results answering your questions directly instead of sending people to the sites that invested in the work.

The thing with the Times is that a large number of people come to it directly, without a newsfeed or search result in between. They type the address into the browser or go to their bookmarks, they open the app out of habit, they go straight to Wordle when they wake up. The free games, the bundle and video are all real, but none of them is rare, and some of it is overstated. Plenty of companies give away a free layer or sell a bundle, but almost none have a direct relationship with millions of people who show up without being sent.

This direct relationship is exactly what the AI companies are designed to capture. And that’s where this turns into a story about power.

OpenAI, Microsoft, Anthropic and Perplexity built systems that sit between a reader and the newsroom, so the reader gets what they came for and never has to make the trip. Perplexity and others answer the question without a click. You’ve seen how Google does the same now. Media companies still pay for the reporting, but the AI model racks up the impressions.

All of this happened without any kind of public process. There wasn’t a hearing or industry input. These companies looted the archives of the entire press (and others) and let everyone find out after the fact. The only place left to fight back is the courtroom, and only publishers with real money could afford to act.

Levien and the Times have spent $20M and 2.5 years in court over the scraping. These AI companies didn’t take all that work just to scrape it. They’re building something that answers your question for you, so you never have to go to the Times. The AI product will be the place people go, and they want people to trust them the way they trust the Times. But without putting in the 175 years.

Look, the Times will probably be fine and continue to thrive. It has the direct relationship, the 175 years, the lawyers. It’s the “paper of record.” That’s not the part to worry about. We have to look out for the middle players, the niche and local sites and the beat reporters who actually break the news. They get scraped the same way, and they can’t afford to sue the AI companies.

Once they’re gone, AI will still be answering people’s questions. But where will it get the information to answer them?

Leave a comment

On May 20, Eversource told state regulators that it wants around $503M more a year from electric customers in Connecticut. 11% on an average bill and 13% on a typical household. The full application will come in a few weeks and PURA then has nearly a year to make a decision.

I wrote last week about the other number, the nearly $15M Eversource reported for its CEO pay in the latest proxy. This, as it asks customers to pay more. The company is short on trust when it needs it the most.

Subscribe now

Your bill is really two bills stapled together. Supply (the electricity itself) is priced by the market. Delivery is what we pay to move it from point A to point B. It includes distribution, transmission, public benefits and more. This case hits just one of these: distribution, which is the poles, wires and crews that Eversource owns and operates.

Here’s what it looks like on a real bill. My bill ran north of $500 for multiple months last winter. Put 13% more on that and it runs me around $65 more a month. Yours will be different. Take 13% of your annual total and you get a rough estimate. A percentage hits a big bill harder than a small one and Connecticut bills run high in the winter.

Eversource’s best argument should be mentioned too. It hasn’t had a full rate review since 2018. It says it’s invested over $3.3B into the grid since then and hasn’t recovered around $600M of it. The costs are real, too: labor, equipment, storms, grid work.

But how much of it belongs on our bills, and whether the company has earned the benefit of the doubt, are different questions.

In 2024, Eversource said it would cut around $500M in Connecticut investment because it disagreed with the regulator’s (PURA) rulings. And now it wants customers to fund grid investment.

A company that threatened to pull back spending after losing fights with regulators shouldn’t be shocked when people have questions after the $503M ask. It’s what you’d ask of anyone who wanted that much money from you.

Here’s the thing press releases miss. This is Eversource’s first rate case in 8 years and it lands in front of a board that looks nothing like the one Eversource fought for years. Marissa Gillett chaired PURA starting in 2019 and built a record of cutting rate hikes. Eversource and Avangrid sued her, and she resigned last October.

Since the new regulators took over, PURA has already approved a $65.9M increase for United Illuminating, the state’s other utility, over the objections of the Attorney General and Consumer Counsel. AG Tong says Eversource waited until it "ran their chief regulator out of town" before asking.

Nobody knows how PURA handles a $503M case, but the early returns favor the companies.

I’ll watch this one closely and come back with what actually matters: what Eversource is really asking for, what PURA makes of it and who pays if Eversource gets what it wants.

The full application is due mid-July. Until then, $503M is just the opening salvo.

Leave a comment

The number in the request is $503M a year. This is the additional money that the Connecticut Light and Power Company, which is what Eversource calls itself when it sells you power here, says it needs for its distribution system. The company estimates that this would raise total bills by about 11% across all customers and 13% for a typical residential customer. The distribution line itself, the part this is really about, would climb roughly 35%. These new rates are proposed for next year.

Subscribe now

This is the first number. There’s also a second number in a different filing.

The proxy statement Eversource filed with the SEC this spring lists CEO Joseph Nolan’s 2025 comp at $14.9M. It also shows a lower adjusted figure of $13.5M, a number you’ll see in some coverage, but Eversource uses the larger total to actually calculate its pay ratio. It’s 104x the median worker’s pay. In 2025, the parent company reported $1.69B in net income.

This information isn’t coming from a critic, it’s from Eversource’s own words.

The CEO’s comp isn’t part of the rate request, and that’s a fair point. The company says that no salary or bonus from its CEO, CFO or president is included in the $503M ask. Even if the CEO worked for free, the shortfall would still be there. It’s a fight over how much it costs to maintain poles, wires, substations, etc.

So the CEO salary doesn’t explain the rate increase. This is 100% true.

But it still matters. Residents who watch this company pay its CEO $15M aren’t going to take a 13% increase lightly. They want the math and Eversource needs to show it.

This is the honest view of the argument. The $15M isn’t causing your bill to go up. It’s the reason Eversource is short on trust at the exact moment it’s asking for a lot of it.

Subscribe now

Now the request goes to PURA (the Public Utilities Regulatory Authority), a board recently rebuilt over the last year. Governor Lamont nominated a bunch of new members and named Thomas Wiehl as the new chair. The law gives them the power to make the call about the increase.

PURA won’t be ruling on whether the $15M salary is excessive. It’ll be deciding how much of the $503M request is justified: which investments matter, what the company is allowed to earn and what we actually pay.

A lot of the coverage misses that nothing will happen fast. Eversource will file its full application around July 14. State law then gives PURA up to 350 days after the effective filing date to make a decision.

Some quick napkin math and it seems like a resolution by next summer, with new rates proposed for July 1, 2027. This won’t be settled this year.

As we wait, the state is pulling levers that move faster than a rate case. Connecticut’s agencies filed a complaint at FERC on June 11 to stop extra revenue Eversource and United Illuminating collect on their lines. It was a bonus on top of the profit they’re already guaranteed.

And on May 1, the public benefits charge flipped from a cost to a credit, which will hold through at least September.

These levers will lower your bill before PURA ever rules on the rate hike. But they don’t affect the $503 million ask.

That fight belongs solely to PURA and it runs into next summer. They’ll decide how much of Eversource’s request is justified and how much of it hits your bill.

The $15M comp won’t appear anywhere in that ruling. It sits in the back of the room while the decision gets made. A company that pays one man that much, in the same year it asks for $503 million more, doesn’t get the benefit of the doubt. A lot of people will be watching the math.

Leave a comment

Recently, I started to do something slightly unhinged. I built a simulator based on my assumptions about the field, calibrated it against live market prices and ran the 2028 primary 20,000 times.

This is math being applied to my opinions. It’s not a forecast. It does two useful things: shows which of your beliefs actually moves the outcome. And it keeps you honest.

What the machine told me was pretty simple. This race won’t really be a competition of the politicians who end up running. Personalities will have an impact, but this year the candidates will be more downstream than in years past. It’ll be a contest between four coalitions.

There’s the Restoration pool: the Black political establishment, party loyalists, the South Carolina firewall: Kamala Harris, Raphael Warnock, Jon Ossoff.

The Populist-left pool: young voters, the movement, anti-oligarchy. AOC, Ro Khanna, Chris Murphy, James Talarico, Graham Platner, Shawn Fain.

Then there’s the Electability pool, which is quite large this year: suburban moderates and regulars who care most about winning: Gavin Newsom, Josh Shapiro, JB Pritzker, Andy Beshear, Mark Kelly, Mayor Pete, Rahm Emanuel and more.

Subscribe now

And the Outsider pool. Jon Stewart? And someone so out of nowhere that they’re not listed here.

This machine asks what would have to happen for one candidate to get ahead and ultimately win the nomination. What variables and conditions need to be set for X or Y to win the nomination?

I’ve identified 5 variables that decide almost everything.

1. Does AOC run? She’s the center of gravity for the Populist-left lane. If she runs, she likely locks up that pool and the race collapses into a binary choice between her and the Electability frontrunner. If she passes, the populist energy needs a home and who knows what happens. AOC running produces a more predictable primary in my simulations. AOC passing produces chaos. See what breaks if she passes →

2. Who votes first? The DNC will set the early-state order over the next weeks and months. It will be a massive decision that dramatically impacts the outcome of the nomination fight. South Carolina could help candidates in the Restoration pool and could help Kamala Harris or her heir. New Hampshire helps Mayor Pete and others in the Electability pool. Nevada, dominated by unions, helps the Populist-left. The campaign event happening right now is happening inside a rules committee. Watch the rules committee pick the nominee →

3. Does Gavin Newsom survive scrutiny? Newsom is the Electability pool’s frontrunner (and overall frontrunner). He’s pretty durable against every calendar decision but fragile under fire. He’s also the only leading candidate that has no kill switch. There’s nothing on the calendar that can stop him from running before voting starts. He’ll be taking fire from everywhere and when he slips in the model, the Electability pool transfers to Pritzker. Put Newsom under fire →

4. Will Georgia go early? Jon Ossoff is the Restoration pool’s swing-state weapon but he’s gotta win re-election in November first. If he romps to victory and then the DNC hands Georgia an early slot, the party will have built a draft-Ossoff machine while he still says he’s not running. Not a bad spot to be. Build the draft-Ossoff machine →

5. What’s 2028 about? Affordability and anti-oligarchy vibes arm the Populist-left. In the model, AI and the future of work arm Ro Khanna specifically. Fighting Trumpism arms the Electability pool. Make it an AI election and watch Khanna move →

The simulator immediately changed three of my beliefs. The worlds where AOC runs are the orderly ones. The worlds where she doesn’t run are the volatile ones. Because the Populist-left group has no obvious second person. The single most disruptive setting in the model is the world where the Electability pool doesn’t consolidate. This would be the opposite of 2020, when the establishment candidates withdrew en masse and endorsed Joe Biden. Across every reasonable variable, the chance that the nominee is a person who’s not on anyone’s list remained stubbornly large.

Now let’s look at the candidates and how their outcomes play out.

Gavin Newsom:

My read. He reads like Hillary in 2008 and 2016. Will he peak too early? He’s seen by the left as too moderate, the center as too left and by the country as just left. He’s extremely talented, well funded and in a weak field he could coast the way Biden did in 2020. He could cut deals with the Populist-left as Biden did. He was seen getting lunch with Elizabeth Warren a few months ago. That lunch, and the leak, was not an accident.

The machine’s read. He’s the Electability pool’s frontrunner and my first model handed him the race, because I had quietly given him an edge in every scenario: he’s the foil if AOC runs and could coast via plurality if she doesn’t. He’s everyone’s plausible second choice but almost nobody’s first love. The one real, big variable is scrutiny. I asked a single question (that you can toggle), does Newsom hold up under fire, his number fell from 29% to the low 20s.

AOC:

My read. The best political athlete in the party. I don’t know if she runs, and that is the point. Schumer’s seat is up in 2028, the same cycle, and the Senate is the “long game” move. Going for the White House is the move you make when the lane is open, and the Obama lesson, which I’m sure she’s familiar with, is that you go when the moment exists.

The machine’s read. Her entry hands Newsom his foil and collapses the race into a binary. Her absence leaves the Populist-left wide open and makes her endorsement worth a lot. The machine really likes AOC.

Jon Ossoff:

My read. People compare his speeches to early Obama. I see it, but differently… to me it’s imitation. It feels like he’s watched old YouTube clips on a loop. In my eyes, he’s a VP option and a remarkable public figure, not a POTUS contender.

The machine’s read. If he wins big in November and the DNC rules committee hands Georgia an early slot, the board will be moving in his favor. My imitation gripes won’t matter if things fall into place.

Kamala Harris:

My read. Harris is the Restoration pool’s main player. I think she’s a poor fit for the moment: she’s shaky on her feet, gets easily flustered and is prone to saying the wrong thing. But she was the Vice President, has universal name recognition (she’s filling arenas) and is the strongest in the field with Black voters, the constituency that decides this primary.

The machine’s read. She’s the most calendar-sensitive candidate in the simulation. She leads in 15% of worlds where South Carolina goes first and only 9% under a traditional calendar.

Pete Buttigieg:

My read. Mayor Pete is the retail end of the Electability pool. People know him and they know he’s smart… but that’s not enough. He needs a massive breakout moment where the country connects to him in a new, authentic way.

The machine’s read. His number nearly doubles under a retail-state calendar (New Hampshire, etc).

Josh Shapiro:

My read. He’s the Electability pool’s strongest pure talent. He’s got a great record in Pennsylvania but is probably too moderate for this electorate. I still think he’s underrated in the markets. But the consensus number two has no launchpad. Pennsylvania isn’t applying for an early slot, he’s no New Hampshire retail creature, and he has no claim on South Carolina.

The machine’s read. He needs Newsom to go away and for this election to be about normalcy.

JB Pritzker:

My read. I think he’s the Electability pool’s most underrated figure. He’s quietly built a machine, spending 2 years funding ballot initiatives across the country.

The machine’s read. He’s the cleanest substitution for Newsom.

Ro Khanna:

The Populist pool’s sleeper. He’s apparently being advised by Ron Klain and is staffed with Bernie 2016 people. He’s always four steps ahead. If the nominee is someone surprising, it could be Ro.

The machine’s read. He’s the most conditional candidate in the field. He needs AOC not to run and for AI / the future of work to be the top issue.

Chris Murphy:

My read. He’s my senator, so read this with that disclosure attached. He keeps surfacing in the model and it’s not because I put my thumb on the scale. He’s merging the Populist-left pool with the Electability pool from inside the Senate, with a strong anti-oligarchy message that breaks through on traditional and new media.

The machine’s read. In the worlds where AOC stays out and the race turns on affordability, Murphy does very well.

The wildcards:

If AOC stays out, the Populist-left energy goes looking for someone. It could be Jon Stewart, the Outsider pool’s loudest option. But for 25 years he’s said over and over again that he has no interest. It honestly could be the lightning rod Graham Platner. If he beats Susan Collins big in November, the Populist-left coalition could morph behind him.

What the machine taught me: I figured out the market has hidden power rankings. If you divide each candidate’s nomination price by their probability of running, you get what the money thinks about them if they enter. Newsom 29, AOC 20, Ossoff 15, Harris 13, Jon Stewart 9.

Also, 1 in 5 of the simulated primaries ends in a contested convention. It’s going to be a historically large field.

The point of this is to show that the candidates are downstream of three things: AOC’s decision, the calendar, and what the country decides 2028 will be about.

Democrats still don’t agree on why they lost in 2024. Whoever finds that answer, and provides a vision for where the country should go, will win. That answer doesn’t exist yet.

The machine doesn’t predict who wins. But what’s cool about it is that it shows which of your assumptions actually change the result. Have fun with it.

The 2028 Machine is live here.

Leave a comment

It’s called the Connecticut Citizens’ Assembly on Property Taxes, led by Comptroller Sean Scanlon’s office and the CCM (Connecticut Conference of Municipalities), in partnership with Yale and UConn.

The final panel was selected this week and the state used the lottery to pick them. At CT Lottery headquarters, officials loaded numbered balls into the machines they use for the nightly draws. UConn’s team mailed 240K letters at random and around 2,300 people responded showing interest.

A second survey verified who was actually around to do it and 1,570 remained. From that they built panels of 110, balancing homeownership, geography and whether people have kids, which generated 50 million possible combinations. Almost 10K of those matched the state’s demographics and 40 were randomly picked as finalists. Each got a ball and the machine ran for 30 seconds. It spat out number 29.

So panel 29 is the assembly, with 110 people all in. 100 members and 10 alternates just in case people drop out. The 100 members will meet over 6 sessions this summer with recommendations due in the fall. Assemblies like this have run in the US before. Petaluma, California used one in 2022 to figure out the future of its fairgrounds. The city adopted the plan. But nobody has tried this at the state level and on a question this large.

And they’re trying to solve a hard problem. Connecticut bars its towns from taxing income and local sales. They live and die on the property tax, which covers around 70% of the average town’s revenues. What a homeowner pays depends on the town’s grand list, the total value of everything taxable within its town limits.

Hartford’s grand list is thin, so it sets a high mill rate to collect what a wealthier town can collect at a much lower rate. Back in 1977, in Horton v. Meskill, the state Supreme Court ruled that the way Connecticut’s schools were funded violated the state constitution due to disparities tied to local property values. The legislature has been tinkering with the formula ever since without touching the dependence part.

Subscribe now

Over the last 50 years, every property tax reform runs into the same question: who’s the loser? More aid from the state equals higher spending. Revenue sharing means some towns get less than they get now. Local-option taxes just shift the burden elsewhere. The state has spent decades arguing over tradeoffs and hasn’t settled on anything.

The assembly’s job isn’t to find a solution that hasn’t been debated already. It’s to figure out which tradeoff the rest of us are willing to live with (or what a representative group of residents is willing to live with). This is the case for an assembly. These residents aren’t running for office and have no constituencies to protect. They weigh the tradeoffs out in the open.

The idea didn’t start in Hartford. It started with Hélène Landemore, a Yale political scientist who for years has argued, most recently in her book Politics Without Politicians, that randomly selected citizens can handle hard questions we usually hand to professionals.

They just need time, good information and a structured process to deliberate. Random selection protects diversity of thought, which elections can filter out. Connecticut is now her largest American test.

This method is older than the country. UConn professor Michael Morrell, who designed the selection, highlighted how the ancient Greeks selected citizens by lot to serve on councils and juries.

Connecticut’s design reads like an answer to the hardest lesson from Landemore’s research. In Iceland, ordinary citizens helped draft a proposed constitution that won broad support in a 2012 referendum. Parliament never enacted it. The process had public legitimacy, but not enough support inside the institution responsible for solidifying it into law.

This version runs through the system on purpose. It is being pushed by the comptroller’s office and backed by the organization that represents the state’s towns.

Comptroller Sean Scanlon has promised to carry the assembly’s recommendations into the legislature. At the drawing, he was blunt about why previous efforts to address property taxes died: legislators feared that if they cast the hard vote, the public wouldn’t have their back. This is built to close that gap. A recommendation from 100 ordinary people gives legislators something property tax reform has never had: cover.

The money is private, not state money: It’s being financed by a mix of Connecticut foundations, civic groups and individual donors.

Citizen assemblies don’t write the actual laws. Their value is narrower than that and real. They shed daylight on the tradeoff and make a room of strangers own it. If 100 strangers can crack the code on property taxes, they’ll hand Connecticut a template for how to tackle more issues that the legislature has let linger for decades.

There's another wrinkle I'll dig into later: organizers say they'll use AI to connect the assembly's deliberations with more than 10% of the state's population.

More to come on this. I’ll be following the sessions over the summer.

Leave a comment

The most interesting sentence from the bill didn’t make it into the law.

The bill that came out of committee this spring laid out something simple. If an AI companion harmed you, you’d be able to sue. Parents could sue if the user was a child. Families could recover actual and punitive damages, plus the state could add civil penalties up to $25K per violation. We know it was in the bill by March 3, because that’s when the industry wrote a letter about it.

Big Tech didn’t want that in the bill. NetChoice, the trade group for Google, Meta and others, told the legislature in March which part worried it most: the private right of action for actual and punitive damages, which their letter said “ensures that operators will implement the most restrictive age-verification measures available rather than risk liability.”

Through early April, the lawsuit was still in the bill.

Then it was gone. The Senate adopted the amendment on April 21 and passed the amended bill on the same day. The lawsuit language was removed and the kids' section now says violations "shall be enforced solely by the Attorney General."

Then the House concurred on May 1, and Governor Lamont signed it into law a few weeks later. The word “punitive” was nowhere in the final product.

Under the new law, essentially no one can directly sue an AI company.

If a chatbot manipulates you and you’re an adult, you can’t sue under this law. If a machine screens you out of a job and nobody tells you, you can’t sue under this law. If a company strips the watermark off synthetic content, you can’t sue.

And now: if an AI companion hurts your kid, this law doesn’t allow you to file a private lawsuit. Your only remedy under this law is the state Attorney General.

A family in Connecticut can still sue the old way: product liability → negligence → plus years of fighting in court → maybe a sealed settlement at the end.

Everything will now run through the state Attorney General’s office. A budget analysis done by the state says the office needs 7 new hires to handle it: 3 more lawyers, 2 IT analysts, a program manager and a paralegal, about $483K in the first year.

This is the enforcement plan for the entire AI industry in Connecticut.

Why does one sentence stand out so much? Because regulations tell a company what it can’t do. Lawsuits decide what happens when it does it anyway.

Subscribe now

Look back to January. Google and Character.AI settled lawsuits brought by parents in Florida, Colorado, New York and Texas. The Florida case was a mother whose 14-year-old son died by suicide after months chatting with a chatbot. The terms of that settlement were never disclosed.

Compliance isn’t cheap, but the expense is limited and very predictable. These tech companies have the cash to budget for warning labels and age checks.

They hate litigation, though. It means discovery, chat logs read aloud in courtrooms and a jury deciding what the company knew and didn’t know.

The cases just disappeared. No trial, no verdict, no precedent, and the question of whether these companies are responsible is still open.

And no federal law will fill this gap. The top bill in Congress belongs to Connecticut’s senior senator. Senator Blumenthal spent 20 years as the state’s Attorney General suing companies on behalf of people who couldn’t afford the lawsuits themselves. His AI bill carries criminal fines up to $250K per offense. It doesn’t put a family in front of a jury.

Other states went the other direction. California’s companion-chatbot law lets families sue. Oregon’s law has no attorney general enforcement at all, only private lawsuits. Connecticut had that sentence in its bill this spring, but took it out.

NetChoice’s letter was sent on March 3. The lawsuit came out of the bill on April 21. We don’t know for sure that one caused the other. But we know the order things happened.

If an AI companion hurts a child in Connecticut, the family is back where the settled families started.

For 7 weeks this spring, the bill said something different.

Leave a comment

For about 20 years, Connecticut hid about a billion dollars of spending inside your electric bill. A 2023 bipartisan law, which Fazio co-authored, forced the utilities to print it as its own line: the public benefits charge, which has run $50 to $60 a month at peak on a typical Eversource bill, close to a fifth of what you pay.

Bringing that into daylight was a genuine public service. Then he built a campaign promise on it. Fazio says he’ll eliminate every hidden tax in your electricity bills and cut rates 20%. His math holds: Getting rid of it would lower the line you pay to the utility.

Here’s the number he won’t give you.

That line isn’t one thing. It is about more than 50 programs: a long-term contract to buy nuclear power, energy efficiency, clean power procurement, help for families who can’t afford their bill and the unpaid bills the rest of us have to cover.

Hardship programs and unpaid bills make up nearly a quarter of Eversource’s public benefits charge and 45% of UI’s. That’s not a subsidy you can just switch off. It’s aid to households that can’t afford it and debt already owed. You can’t wipe it off a bill, you have to move it somewhere else.

Watch out for the word “eliminate.” Fazio’s actual plan is to end some programs, phase out others, and move the ones he likes to the regular state budget. Where they’d be up for debate every two years.

Fazio’s supporters will say moving the charge to the income tax is fairer, and they have a point. But fairer isn’t gone. Most of the 20% isn’t being cut, it’s just being moved. The money has to come from somewhere.

The real question isn’t the 20%. It’s who pays for it once it lands in the budget.

Subscribe now

He hasn’t put any numbers to it and that is the hole at the center of the plan. To see what a billion dollars means in the state budget, look at the fights in Hartford over just tens of millions: $40 million in emergency aid in 2025 for a special education crisis, a Medicaid rate increase cut from $75 million to $15 million.

Fazio wants to move close to $1B into that fight. And here’s the catch. Connecticut has spending caps, the fiscal guardrails from 2017 that Fazio himself has championed for years. A billion dollars of new spending can’t slip in under those caps. Something has to give, but he won’t say what.

There’s one cut he does name, and it proves the point. Last year he co-authored Senate Bill 4, the omnibus energy bill signed as Public Act 25-173, which cut the charge by about 10%.

A clean example is the nuclear credit. On May 1, bills dropped around $30 a month for Eversource customers and $34 for UI, because the public benefits charge fell so far that it flipped into a credit. The driver was CT’s fixed-price contracts with Millstone and Seabrook, which have returned about $450M to ratepayers over the last 2 years.

Fazio wants to keep those contracts. He likes nuclear power, and that’s good. But that credit reaches customers through the very charge he wants to get rid of. If the public benefits charge goes away, does the $450 million windfall flow back to ratepayers, or does it land in the General Fund (the state’s main checking account)? His plan doesn’t say.

So here’s the ask… Name the programs that you’d cut. Plainly say that the charge is being moved, not eliminated. What will happen to the nuclear credit? What number goes up when this one comes down?

In a June 2 op-ed introducing himself to Connecticut voters, Fazio repeated the promise. He’ll eliminate every hidden tax and cut rates 20%. He had plenty to say about the laws he’s passed and the taxes he blames on Lamont. He didn’t say what his plan would cost.

Fazio calls it eliminating the charge. It’s really moving it, and he still won’t say where.

Leave a comment

Fazio's campaign did not respond to a request for comment.

Public Act 26-15, signed earlier this month under the bland title An Act Concerning Online Safety, runs 74 pages. It regulates frontier AI models, hiring tools, AI companions for minors, synthetic media and subscription chatbots.

*See below.

Subscribe now

Almost every part of the bill that has teeth runs through one desk.

An asset manager in Stamford using AI to screen every job applicant before a human even sees a résumé. A 14-year-old in New Haven talking to the same chatbot after midnight. A synthetic video or image used in a campaign mailer without the digital watermark that the law now requires.

Under Public Act 26-15, one man gets to decide what to do about them.

Connecticut isn’t the only state to centralize AI enforcement within the AG’s office. Colorado does something similar, as does Texas. What’s unusual about CT is the breadth: one bill and one office covering frontier models, chatbots, hiring tools and AI/synthetic media. California is the outlier the other way, spreading authority across the AG, the new CA Privacy Protection Agency and a handful of laws that let regular citizens sue directly.

Congress has been largely absent on AI regulation. The FTC has been active on AI but mostly on narrow cases regarding deceptive marketing. In this world, CT consolidated power and handed most of it to Tong.

Tong indicated earlier this year that his office already saw AI as part of its consumer protection and civil rights mandate. Public Act 26-15, which state Senator James Maroney shepherded through three legislative sessions to get passed, gives that posture teeth.

The short-term stuff is obvious. He’ll have to figure out what “related to AI” really means when companies start filing WARN notices with the Department of Labor in October. He’ll have to decide whether to go after corporations using AI to hire people or wait and see what cases emerge.

What gets less attention is what happens after Tong. He won’t be AG forever and he’s on the shortlist for any bigger statewide office that opens up. Whoever comes next inherits a portfolio that didn’t exist before May 1.

The effects will show up in CT residents’ lives long before most of them ever hear about “Public Act 26-15.”

October 1 is a big day. It’s when the first WARN notices with an AI disclosure box can start being filed.

A policy team at UConn owes a workforce impact report by January 1. The AI working group’s initial meeting is required by August 31. Watch those moments to see what Tong’s office does… and what it doesn’t.

William Tong has a new job. It’s worth watching what he does with it.

Leave a comment

*Update: the graphic originally said Sec. 6 allows private suits, following early legal analyses. The enrolled text says violations "shall be enforced solely by the Attorney General," and the bill's explicit private lawsuit was removed before passage.

Leo, the first American pope, didn’t pick his name by accident. Leo XIII wrote Rerum Novarum in 1891, the church’s response to the Industrial Revolution and the collapse of the old world labor order. This Leo picked his name to signal that AI is a rupture on that same scale. It’s not another tech cycle in his eyes. It’s a shift in power, work and what it means to be useful.

Big Tech has spent a decade insisting that old institutions were obsolete. Now some of the people building AI are looking to these institutions, like the Vatican, to help govern what comes next.

Olah argued at the Vatican that AI development needs scrutiny from governments, religious leaders and civil society. He spoke frankly about the tension inside the AI labs: commercial pressure, political pressure and the reality that companies racing to build the future may not be equipped to govern it.

He even brought up job displacement directly. This isn’t usually how Silicon Valley talks.

For most of the last decade, the industry has treated religion, government and the humanities more generally as obstacles to innovation and growth. They need to move fast, scale globally and let the regulators catch up. Just like social media (didn’t that turn out well?).

Now some of the top labs are publicly admitting that the market alone won’t be capable of governing AI.

The encyclical’s sharpest line also happens to be its plainest. Leo XIV writes that “the promise of automatic general prosperity often proves illusory.”

Subscribe now

AI is moving too quickly for a trickle-down version of it to happen on its own. Someone has to decide who will benefit. Someone has to decide who will absorb the disruption. Does efficiency matter more than stability or dignity?

Those are political and moral questions. They’re not being raised only by critics outside of the room. Now, the Pope and the Vatican are in the room, working with one of the leading figures inside Anthropic.

The takes are already sorting themselves out. Some will argue the Pope doesn’t understand technology or even consciousness. Others will say the real risk is the government, not the companies. Both critiques are missing the point and not talking about what happened in the room with the Pope and Chris Olah.

The consciousness debate is easier to have than the power debate. That’s why so many critics prefer to use that line.

The Vatican is treating AI like a societal shift and DC still mostly treats it like a tech story.

Just this week, President Trump pulled back an executive order that would’ve created government testing for advanced AI models after pressure from allies like David Sacks.

The Pope is talking about labor, power and human dignity. He gets it.

Washington is still listening to the people who stand to profit from it.

Leave a comment

None of these companies said that AI took the jobs. On earnings calls and in press releases, execs talked about efficiency, automation, restructuring and “doing more with less.”

This is the language of the AI era, whether or not anyone was willing to go out and name it.

The pattern keeps going beyond the headlines. Connecticut’s unemployment rate grew to 4.8% in March, up a full point from the prior March. Professional and business services, the category that covers most of office work in the state, lost 1,800 jobs in March alone. Companies aren’t having mass layoffs. They’re just not hiring back and the career ladder is being shut down one rung at a time.

Meanwhile, almost 11K job postings in Connecticut list AI skills as a requirement. That’s a 40% jump in one year. If you’re 45 and working in a West Hartford office park or a tower in Stamford, your job will probably still exist next year. The question is what it looks like in 5 years.

A few months ago, researchers at Tufts published a study that ranked the states most exposed to AI displacement. The top of the list wasn’t surprising: DC, Massachusetts, Virginia… states full of financial analysts, lawyers, developers and consultants.

Connecticut isn’t at the very, very top but the industries the Tufts authors flag as the most exposed (finance, information and insurance) happen to be Connecticut’s spine.

The states that won the knowledge economy over the last 50 years are the states most exposed to what comes next.

For the last 2 years, Connecticut politics has been all about affordability, electric bills, property taxes, health care premiums, housing and more. The gubernatorial race will be fought, almost entirely, over what your monthly bill looks like.

But all of this talk assumes one thing: that people will keep making the same income. What happens if the income side changes too? That assumption is starting to break.

Subscribe now

This is the version of affordability that nobody is running on yet. The next fight is about whether you still have a job, not what your bills cost last month.

Voters aren’t waiting for Washington for answers. Polling from Blue Rose Research asked Americans who they trusted to protect their communities from the downside of AI data centers. The choices were the White House, Democrats in Congress, Republicans in Congress or governor. The governor won 66% of the head-to-head matchups and none of the federal options cleared 50%.

The conventional wisdom hasn’t caught up to this yet. Ryan Fazio, the Republican nominee for governor, only talks about utility rates. The legislative debate is all about the public benefits charge on electric bills. These are all essential and important issues. But no one is making the case that AI is coming for the work itself, not just the cost of living that it supports.

But voters are already there and are waiting for someone to say it out loud.

Last month, Connecticut passed one of the country’s first major workplace AI laws. Starting in October, companies using AI to evaluate, rank or terminate workers will now face new transparency requirements.

Right now, everything can be hidden behind corporate language. This new law forces a paper trail.

Disclosure is a big step forward, but it’s not protection. New Jersey recently passed a law that charges companies fees when they replace workers with AI. The state then uses that money to fund apprenticeships and retraining programs.

Connecticut now has transparency, but the harder question is what comes next.

The state has advantages most states don’t. Wealth, education levels, strong institutions and a governor who actually understands technology. The state legislators moved earlier on AI than most states.

But the same economy that made Connecticut wealthy also made it unusually dependent on the type of work that AI is squeezing first. The disruption isn’t going to arrive all at once. It’ll show up in hiring freezes, quiet restructurings and in WARN notices that actually have to say the word “AI” out loud.

On October 1st, the AI disclosures will begin. What Connecticut does with them is the real question.

Leave a comment

At Google I/O last week, the company revealed one of the biggest rebuilds of search in a generation. The search box will no longer just be a path to the web. It’s becoming the web: conversational answers, agents and generated interfaces that live inside Google.

For users, this may seem convenient. For publishers, it’s something closer to a platform obituary.

For 20 years, Google search has been the backbone of the internet. Publishers didn’t love every part of the deal, but the deal worked. You built a website, you wrote stuff people wanted to read, people found you on Google.

That was the deal.

A decade ago, the typical content site would get 30 to 60% of its traffic from Google search. The rest came from direct visits and social, mostly Facebook. The ratio was the foundation of nearly every business model in digital media.

I’ve seen entire newsroom strategies built around this math. Hiring and content plans, weekend coverage, SEO teams, live blogs, commerce, all of it started with the same assumption: Google, the heart of the open web, would keep sending readers.

Then Facebook cut off the social fire hose.

After the 2016 election, Facebook gradually stopped sending publishers traffic. The reasons shifted year to year. First they said it was manipulation of link posts leading to election interference. Then it was the pivot to video, supercharged by metrics Facebook later acknowledged were inflated, but not before publishers laid off thousands of writers to chase it. Then it was algorithmic deprioritization of news articles that simply was never reversed.

By 2020, publisher referral traffic from Facebook was down 90% on average. It’s gone down since then, too.

Subscribe now

The social platforms are now designed to keep you inside them. Their incentive is to never let you leave. Now, the same thing is happening with Google and it’s a much bigger deal.

At Google I/O, the company described its new search update as the most significant change in 25 years. Users will get conversational answers, agents and new generative interfaces that will live inside Google. The traditional results page with blue links won’t be the product. Google is moving on.

Google will say this is better for users. In many ways, it will be. But when Google gives people the answer directly, fewer people click on the links to the sites that did the real work in the first place.

The web numbers for publishers were already brutal. Chartbeat covered 2,500+ news sites showing Google search referrals were down 33% globally year over year and over 38% in the US. Right now, 69% of searches are zero-click searches. The number is climbing.

Google Discover, the feed you see when you scroll down on Google mobile, is a remaining fire hose for many publishers. It’s not much better. It can deliver huge traffic but it’s brutal to build around. You’ll have a story that should work that does nothing. A throwaway post will catch fire. It rewards luck in the algorithm over audience loyalty.

One more detail worth sitting with: around the same time Google was rebuilding search in a way that would hollow out the publishers that feed the web, Sergey Brin was back in Alphabet’s orbit, leading a coding sprint at DeepMind.

Brin has relocated to the Nevada side of Lake Tahoe amid California’s proposed and likely doomed billionaire tax fight, which he donated $45M against. And now, by Donald Trump’s own description, he is dating a “really wonderful MAGA girlfriend.” Brin’s made no public case or comments for how Google’s AI search era is supposed to work for the publishers that built it.

While publishers are figuring out what happens next, one of the guys shaping the future of the internet is living in his completely different world.

There was no public process or industry listening tour. Just a handful of executives in Mountain View made the call.

So what happens now? Some publishers will adapt, others will fold. The ones that win will diversify by owning more of the relationship: newsletters, SMS, paid memberships, social content and franchises that people seek out, repeatable video formats, live events and other ways to activate the audience that a publisher already has. This is where innovation is right now.

Publishers will still take Google traffic. But they can’t build a business assuming that Google will keep sending readers. The old deal is done.

Leave a comment

(Video courtesy of Bleeding Heartland)

It sounds fake and corny when you first hear about it. Like something from an Aaron Sorkin movie or a consultant’s daydream about restoring civility. But it’s very real and it’s now happened over 100 times in 6 months across the state of Iowa.

Rob Sand is the only Democrat who holds statewide office in Iowa, which has become very red over the last few decades. He’s held town halls in all 99 counties. The events aren’t flashy. But the opening itself says a lot about how Sand views politics.

Subscribe now

Most politicians begin with the assumption that the electorate is permanently divided into enemy camps. They must energize their side, inflame and depress the other side and keep people angry enough to stay engaged. And much of the time this works electorally but it creates a culture where Americans treat each other like mortal enemies.

Sand is reversing this dynamic, or at least interrupting it for an hour.

The point of the singing isn’t that Republicans and Democrats suddenly agree with each other. It’s that before anyone starts arguing about taxes or Trump or schools or whatever else, everybody in the room is asked to acknowledge that the other people in the room are still Americans too. They all belong. It’s an incredibly low bar. Right now, it feels strangely powerful.

There was a time when life in America required a lot more interaction between people who disagreed politically. There was more activity in churches, town papers, civic clubs, Little League and town meetings. A lot of these institutions or associated activities have moved online, where it’s easier to engage in conflict and harder to coexist.

What Rob Sand is doing feels old fashioned in a way that catches a lot of people off guard, in a good way.

Share

And politically, it’s working. He’s getting a lot of Republicans at his events. Sand is pragmatic and local papers describe him as “non ideological.” He’s raised millions after quickly entering the race. He doesn’t talk about Iowa voters as if half the state is morally contaminated and deplorable, with the other half needing to be permanently mobilized against them.

Americans keep saying they want less polarization but most of the incentives reward the opposite. The people who stand out now are the ones who are willing to buck this trend and act differently in public, not just talk differently online.

Rob Sand is on to something big: democracy works better when people see each other as neighbors first and partisans second.

That should be completely normal. Right now, it seems radical.

Leave a comment

“So how is this legal?”

The reason for the question is obvious. John Kissel has been in the CT State Senate since 1993. He’s one of the most senior Republicans in the Capitol. He also happens to be a corporate attorney for Eversource, the monopoly utility that sends electric bills to most of CT.

It seems like the type of thing that should be illegal. Or at least the kind of thing that would trigger a new ethics rule. Eversource isn’t a random private employer. It’s the regulated utility that most Connecticut residents have to use. They lobby the legislature just as much as anyone. They fight with PURA (the state’s utility regulator).

So the Reddit question makes a lot of sense. How is this legal?

The state wrote it this way. Under CT’s ethics statute, a legislator only has a “substantial” conflict of interest when a piece of legislation would financially benefit the legislator, a family member or a “business with which he/she is associated.”

That sounds broad until you read the definition, a “business with which he is associated” doesn’t include a legislator’s regular employer.

And this is the whole ballgame.

If a lawmaker owns part of a company, directs the company or has specific leadership ties to it, that can create a conflict of interest. If the lawmaker just works at the company, the rule gets much softer.

This isn’t my take on it, either. 9 years ago, the Office of State Ethics told then State Senator George Logan that Aquarion, his employer, wasn’t technically a “business with which you are associated” under the law.

They said Logan could vote on legislation affecting Aquarion, as long as the bill didn’t directly put cash into his own pocket or the pocket of an immediate family member. They also said he could vote on budget, bond or tax packages even if they included things specifically benefiting Aquarion.

This is how Connecticut currently handles conflicts. The question isn’t whether your employer has major business before the state. It’s whether the vote creates a direct financial benefit to you personally. This is an extremely narrow standard.

Mark Pazniokas of the CT Mirror had the cleanest take on this in 2020: “Nothing in Connecticut’s narrowly drawn ethics rules bar legislators from voting on matters that help or harm their employers.”

Which brings us back to Senator Kissel.

It appears Kissel has been careful. He’s not on the Energy and Technology Committee, which would be a major conflict of interest. He does typically recuse himself from energy bills, even though he’s not required to do so.

In 2020, when a big storm hit and there was public anger over outages and rate increases, he voted for the utility accountability bill. “This doesn’t benefit my employer and my constituents spoke with one, loud voice, especially after the rates went up.”

This actually matters. This piece isn’t about how Kissel got caught doing something illegal. There’s no ethics finding or investigation that I’m aware of. He may well be acting in good faith.

But this is the crux of the problem. The system depends on him acting in good faith.

Senator Kissel is a senior member of the Regulations Review Committee, which reviews state agency regulations before they go live. That includes regulations from PURA, the agency that regulates Eversource.

So even if Kissel avoids the committee that writes Connecticut’s energy policy, he still sits near the machinery that turns PURA regulations into state rules.

Again, this is 100% legal and Kissel was not the only one.

Subscribe now

From 2017-2020, 3 GOP senators had direct professional ties to Eversource or a subsidiary while serving in the legislature. As noted, Kissel works as a corporate attorney for Eversource. Kevin Witkos, then the deputy minority leader, worked for Eversource as a community relations specialist. Logan worked for Aquarion, owned by Eversource.

They weren’t sitting around killing bills. The point is more basic than that. Eversource had people in the building.

Democrats have their own versions of this too. This isn’t a story about one clean party and one dirty party. Eversource’s bench in Hartford has been mostly Republican, this is also true.

The question normal people are really asking isn’t whether every vote is dirty or whether every lawmaker is corrupt. They want to know if they can trust the system.

I don’t think they can, specifically when it comes to Eversource.

In 2023 and 2024, the company spent around $1.3M lobbying Hartford. Eversource and United Illuminating sued PURA Chair Marissa Gillett last year. The Eversource CEO announced $500M in Connecticut spending cuts the same morning Governor Lamont reappointed Gillett to PURA, blaming a hostile regulatory environment.

Lamont was spot on when describing the campaign against Gillett: “There’s a full-court press to get her out of there… I mean the lobbyists, the publicists, the lawyers.”

I’m not the first person to call all of this out. In 2020, Peter Lewandowski (then director of the Office of State Ethics) proposed requiring lawmakers to file a written disclosure when they voted on something affecting their employer. It didn’t even stop them from voting but at least it was something.

But even that bill died in committee.

Share

Common Cause Connecticut has been trying for years to get the legislature to address this and tighten these rules… but the proposals go nowhere.

Massachusetts has a different standard. Their ethics law treats a business where a lawmaker serves as an employee as a potential conflict. So the thing Connecticut carved out, Massachusetts wrote in.

The Reddit user asked the right question. “How is this legal?” It’s because CT lawmakers made it legal and then they left it that way.

The next time your Eversource bill jumps, you can get mad at the company and the politicians who keep insisting that the system works.

But keep in mind that Connecticut has a system where a state senator can work for a monopoly like Eversource, sit close to the regulatory process and still be following the rules.

The rules are the problem.

Leave a comment

The CEO of the most valuable company on the planet wasn’t on the list.

Jensen Huang of Nvidia had been left off. Reporting suggested it would’ve complicated talks with Xi around AI export controls.

But by Tuesday afternoon, the situation changed. On Tuesday night, Huang was trotting across the tarmac in Alaska, boarding Air Force One as it refueled. Then the president posted on Truth Social that Jensen was onboard.

This image tells you everything you need to know about the relationship between the US government and corporate power in 2026.

Trump landed in Beijing on Wednesday at exactly 8:08pm local time. This timing was chosen because 8 is the luckiest number in Chinese culture.

Subscribe now

He got a warm reception with a red carpet, military band and hundreds of Chinese school children waving American and Chinese flags in carefully synchronized motions.

The president came down the stairs and introduced the delegation like a billionaire at Sun Valley introducing friends at breakfast. And standing right behind them was the man who had caught the flight in Alaska 12 hours earlier, backpack still slung over his shoulder, while Elon stood there in black and Tim Cook in a suit.

Executives have traveled with presidents before, it’s not new. The old separation between political power and corporate power has evaporated. The CEOs aren’t waiting outside the room anymore. In many cases, they’re calling the shots.

In 1938, FDR put Thurman Arnold in charge of antitrust in the DOJ. Arnold treated concentrated corporate power as something to confront, not console.

This version of Washington, DC feels very far away.

Share

Jensen Huang caught Air Force One in Alaska because he’s got business to handle in China.

All over America this morning, people opened utility bills, filled their gas tanks and just kind of absorbed it, because what else are they supposed to do?

One group gets a flight on Air Force One. The other gets told the bill is due.

Leave a comment

For most of the last decade, these weren’t the type of stories you’d hear about Erin Stewart.

You heard about the 26-year-old wunderkind when she won. You heard about the Republican winning in a Democratic city, the daughter of a former mayor, who was blunt and telegenic. You heard that she could be the future of the Connecticut GOP. The New York Times profiled her. She had the kind of coverage that turns local politicians into stars.

But the crack was there if you looked. In 2018, when she made her first moves towards statewide office, she finished a distant second for the lieutenant governor nomination at the GOP convention. The reporters and power brokers pontificating about her rise were never quite the same people voting in Republican primaries.

Last month, the Mirror reported that Stewart had asked New Britain for a partial pension benefit that the city charter doesn’t include. She actually wrote to HR herself: “Based on my 14 years of service, I calculate my eligibility for a deferred partial pension benefit of approximately 35% of my annual salary, payable upon reaching the eligible age of 55.”

The city has no such benefit for someone with under 20 years of service.

When asked why she made the request, she told the Mirror, "Why wouldn't I?"

Subscribe now

The whole aura around Stewart rested on competence. In a state this small, reputations compound slowly over time and collapse all at once.

Connecticut only has 169 towns and maybe 400 people or so who actually run its politics. Most of them have known each other for decades. Stories travel quietly for a long time before they hit the public.

Before Republican primary voters started drifting to the state senator from Greenwich, New Britain already had their own verdict on Stewart. Her hand-picked successor lost to Bobby Sanchez last November. Mayor Sanchez is the one who hired the law firm and released the records.

We’re only hearing about this stuff now because the city she ran for 12 years is no longer run by her people.

Erin Stewart’s argument was competence. Her proof was New Britain. And now New Britain is the source of the questions.

Leave a comment

Last year, the Senate passed a real AI bill with teeth and the House just let it sit there until the session ended. There was no vote, it just disappeared. It’s a pattern you see more than you think.

But today, the House passed SB5 anyway… 131 to 17.

And now it goes to Governor Ned Lamont, who’s expected to sign it.

For the first time, Connecticut actually gets an AI law through both chambers. This is a huge moment.

Subscribe now

The story doesn’t end here, though.

Big Tech is already pushing back. Just this week, a trade group sued the state of Minnesota over a similar law. There’s also talk of a federal preemption from the Trump administration that would entirely wipe out all state laws on this issue.

But today the bill passed. Now we’ll see if it sticks.

More soon.

Leave a comment

The House never even voted on it. It just sat there until the session ended. Then it was gone.

This isn’t a fluke. This happens way more than people realize. And now, it’s happening again.

A few weeks ago, the Senate passed SB 5. It’s a bigger bill this time and you can feel the compromise in it. The edges softened and new pieces were added. It looks like something that’s been massaged, not something that came out clean the first time. But it’s still real and it would change things in a real way.

It’s sitting in the Connecticut House with no vote scheduled. The session ends on May 6.

I reached out to Senator Maroney, one of the architects of the Senate bill, this week to get a read on what happens next. If his office responds, I’ll update this piece.

That’s part of the reality here too. Even if you’re paying attention and following closely, it’s not always clear who’s making the calls or what’s actually happening behind the scenes.

What the bill does is straightforward, once you strip it down.

It creates an AI policy office and an AI workforce academy inside the state government. Essentially, the state is attempting to get its arms around something it knows is already here.

It puts new rules around how AI gets used, especially with kids. Chatbots, social media and the whole blurry category of things where everyone agrees there’s a problem but no one agrees on how to deal with it or where the line is.

Subscribe now

The part that stuck with me most is the hiring piece. The bill treats algorithmic hiring as something that could cross into civil rights territory. It also prohibits AI from interfering with state employee union collective bargaining agreements. This is the kind of thing that shows up in someone’s real-life job search or workplace.

If the bill dies, all of that will go with it. It won’t be revised later or delayed. Just gone.

You can argue that this bill is messy. It is.

You can argue that it’s trying to do too much. Maybe it does.

But it’s also one of the more serious attempts that Connecticut (and most states) have made to deal with this at all. And right now, the pattern looks familiar.

The Senate does the work. Then the House decides whether anything happens.

One thing I’ve learned is that most bills don’t get voted down. They just never come up at all.

Share

So the real question isn’t really what’s in SB 5. It’s whether it gets a vote at all.

This decision sits somewhere in the CT House leadership. The Speaker or the people around him. I don’t know who exactly makes that call. This is kind of the point.

It’s also been interesting to look at who showed up around last year’s bill that stalled and failed.

Lobbyists, tech companies, the usual players. More on this to come.

For now, it’s much simpler. There’s a bill and a deadline. And there’s a pattern that says if nothing happens, the bill will disappear.

Update: On May 1, 2026, the Connecticut House passed this legislation.

Leave a comment

That’s a different experience. On one side, OpenAI is laying out how AI will reshape work, prosperity and society. Wealth will concentrate, governments will misuse the technology, etc. It’s all there very plainly.

They actually get it. Pretty swiftly, this went from being a product story and started looking like a power and control story.

On the flip side, Farrow’s reporting includes direct accounts from ex-employees of OpenAI who questioned whether leadership was being fully candid internally at key moments. You can’t really separate this.

In the policy paper, they keep coming back to the New Deal. This is the analogy they want: a big shift, a country adapting, new systems being built, spreading the wealth.

But let’s actually sit with that.

The New Deal wasn’t a conversation or a document. It was a fight against power and entrenched interests. The government stepped in because concentrated power started to distort everything in its path. They built agencies that could actually enforce things. They backed labor and broke up monopolies.

They didn’t ask the OpenAIs of that time to write the rules. This is the part that matters.

Subscribe now

Because what’s happening now is different. OpenAI is essentially saying: we see the risks, we agree that things need to change and here’s a vision for how it should work.

Some of it is legit, specifically the parts where they acknowledge and describe how disruptive this technology will be.

The rest is where it gets cleaner and more manageable. The kind of solutions that don’t box them in. They’re getting ahead of the solutions that don’t inherently benefit them.

If you’re in control of something this powerful, you don’t design rules that meaningfully limit you. You design rules that look reasonable on the surface but benefit you in ways that normies can’t see.

And this isn’t unique to OpenAI, this is how incentives work.

Share

There are many parts of their document that are serious. The Public Wealth Fund idea, not because it’s ready to go tomorrow but because it says something most companies would avoid.

They essentially say that if this works, the upside will concentrate. A few companies will hold a lot of the value. Everyone else gets it downstream.

Same for the tax section. They’re admitting that the current tax system won’t hold if income shifts away from labor and toward capital.

The paper’s safety net ideas are more responsive than what I’d expect, calling for support ramping up when disruption spikes, then pulling back when things stabilize.

This tells me they’re planning for disruptions and don’t expect smooth sailing.

The bigger gap is what they don’t touch.

I’ve spent most of my career inside of the attention economy. I’ve seen what happens when content gets cheaper to produce and easier to distribute. I’ve seen the incentives bend in real time.

AI brings this dynamic to another level. These systems generate text, images and video. This is the raw material of how people understand what’s happening. And now these systems can generate content instantly, at scale.

The policy paper talks about trust, verification and democratic values… but stays at a high level.

We know where this goes. Content gets cheaper… a lot cheaper. It gets easier to target, tweak and flood every platform with slightly different versions of the same message until nobody is quite sure what’s real anymore or what matters.

We’ve seen a smaller version of this with social media. That seemed massive but was a baby step compared to what’s coming.

And everything else they’re talking about in the paper (accountability, oversight, input from normal people) depends on all of us agreeing what’s real in the first place.

This is the part OpenAI mostly skips.

They talk a lot about success and making sure people have access to AI. The words “affordable,” “widespread,” and “something like a right” are peppered throughout.

And this stuff is important… but access isn’t control.

They’ll allow everyone access but still have the power to make all of the decisions: how it’s built and behaves and how value flows. All of this to be controlled by a small handful of companies.

It’s not about access, it’s about who has the power to decide. It’s the difference between being a user and having power.

The “Right to AI” idea blurs this line. It sounds like a public good but it also expands the customer base for the people building these tools. Both things can be true.

Some of the rest feels familiar.

Governance ideas that sound solid until you ask who actually enforces them. Public input that feels more like being heard than having power. Auditing that focuses on only the highest-end systems, which makes sense until you realize that’s not where most of the risk sits.

Nothing in this policy paper is crazy… but that’s almost the point.

Subscribe now

And then there’s where this actually lands. Not in DC but in states, schools, hospitals and municipal budgets.

These places are already trying to figure out what this will mean for jobs, energy and what gets built next.

States like Connecticut are already in the thick of it. Bills moving, Silicon Valley lobbying money showing up, rules getting shaped before most people are even paying attention.

Corporations are spending real money to influence how the AI rules land. It’s not theoretical, it’s the ground game being played right now.

And it’s happening below the level OpenAI’s policy paper was written for.

Share

Sam Altman and OpenAI understand how big this is.

They’re also the ones closest to the center of it. They’re putting the first version of the rules on the table.

But this only works if they’ve built up a certain level of trust.

They’d need to be straight about the risks, follow through on constraints and help design something that doesn’t just entrench them further.

The funny thing is, the argument inside OpenAI (not that long ago) wasn’t about policy. It was a fight about whether the people in charge were being fully candid about what was happening inside of the company.

It’s impossible not to allow this to change how you read everything else.

The New Deal didn’t work because people trusted the right executives. It worked because it didn’t need that.

It built systems that could actually check power. It didn’t ask companies at the center of it to write the rules.

This time, they are.

Leave a comment

Max Stahl thinks you’re asking the wrong question.

I first met Max Stahl in June 2007. We were living in the same dorms in Manchester, NH, part of a group of two dozen Obama for America interns canvassing towns where most people didn’t know who Obama was.

We’d go door to door in Bedford and Amherst (and other towns) carrying supporter cards, competing to see who could get more voters to commit to Obama.

Max always won. He had this money-and-politics pitch that worked like a charm, even then. He dropped out of college to work on Obama’s Nevada operation, eventually went back to school and has never really left politics.

I went into media. We’ve stayed in touch ever since… and money in politics has been the thread I’ve kept pulling on because I think most of our other problems trace back to it.

Everyone knows money runs politics. But almost nobody has a clear idea what to do about it.

Every reform idea in the last 30 years has had the same issue. Disclosure requirements: the money found a way around it. Public financing: popular, bipartisan but never implemented at scale.

Subscribe now

Money in politics is like water through rocks. You plug one hole and it finds another. Always has.

Max has spent 20 years inside this fight. In 2014, he ran a campaign on college campuses with interns making $1,500 a semester and no marketing budget. He had one goal: get presidential candidates to sign a one-line pledge to take big money out of politics.

Hillary Clinton, whose donor network was the very thing they were fighting, never officially signed. But by the end, the Democrats’ platform at the 2016 convention had his language in it anyway.

Max has watched that water long enough to know where the real crack is. When I spoke with him recently, he said something that reframed the Citizen United debate for me.

The conventional take is that CU opened up the floodgates (more money, worse politics).

Before Citizens United, even the most corrupt power centers had real constituencies. When the auto industry or oil companies did well, their workers did well, their states did well. Michigan felt the success when Detroit did. The people in the smoky back room, as Max puts it, “when they did well, they brought a lot of people with them.” It wasn’t some noble exercise, the constituencies forced their hand.

Citizens United changed who was at the table. Now you can have a single anonymous individual, no company, no workers, no factories, no real constituencies anywhere, buying as much influence as the auto industry once could.

You don’t know who these people are. You have no idea what they want. And their interests don’t have to align with anyone’s but their own.

This is why every fix aimed at “less money” keeps failing and won’t work.

So here’s where Max’s landed after 20 years of doing this.

We didn’t ban cigarettes. We heavily taxed them and used that money for the public good. Lottery tickets fund local schools. Sin taxes exist because we decided certain behaviors, while legal, should carry a public cost.

What if we applied the same logic to political donations?

Small donations would remain untouched: you and I give $25, it’s not taxed. But once you start spending big money, a progressive tax kicks in.

Hit $10K and you’re taxed at 100%. $20K, you’re paying 200%. Want to spend $10M on a race? Now it costs you $250M.

Subscribe now

You can still spend the money, you just can’t profit from it.

We’ll never be able to remove the incentive… but we can change the cost. Right now, political spending is the best investment a billionaire can make. A $3.5K max donation to a congressional candidate (“cheap dates” as Max calls them), buys access. Access becomes information. Information becomes a market edge.

Max described a trade placed 15 minutes before Trump announced a “peace deal” with Iran, billions in calls timed to the announcement. The market surged.

You and I can’t make bets like that. The people with the right cell phone numbers can.

Let’s flip this math. Make the cost of buying influence so punishing that the ROI disappears. It won’t be illegal, just expensive enough that it stops making sense.

Where should this money go? Max has two ideas:

1. Fund public financing so candidates without billionaire backing can compete.

2. Send checks directly to people.

Max’s girlfriend was canvassing in Detroit for Kamala Harris in the days before the 2024 election. She knocked on a door in one of the poorest neighborhoods in the city. The person answered the door and quickly said they were voting for Trump.

She asked why: “Trump sent me a check. No other politician’s ever done anything for me.”

People remember when the government does something for them. It’s very sticky and it shapes how people vote.

Sin tax revenue could be a player in this. Not just a funding mechanism but a tangible signal every 2-4 years of how much the game is being played… and who gets cut in.

In 2007, we were 20 years old knocking on doors in New Hampshire for a candidate nobody believed in yet. Back then, Max was already better at the pitch than I was. He’s been refining it ever since.

Our system runs on incentives. This is one of the few ideas that actually changes them.

Not by stopping the money. By making it too expensive to matter.

Leave a comment

What happens when the economy keeps growing… but fewer people are needed to run it?

And I don’t mean in some far off, fantasy, sci-fi sense. I mean it in a very real, measurable way. You can already see it in hiring trends, earnings reports, and how work is quietly being reorganized right now.

Output goes up, revenue holds and over time, the number of people required to sustain that output starts to drift downward. It’s not a hypothetical anymore, it’s a direction.

The discourse on AI is stuck in two places that don’t really connect.

On one side, there’s the optimists who talk about AI tutors for every child, exponentially faster drug discovery and a more productive economy. Some of this will be true.

On the other side, you’ve got the doomers. Job loss, misinfo/disinfo, moving towards a world that no one can define but everyone gestures towards.

Both sides are reacting to real things. But neither is talking about the structure of the economy itself and that’s where it’s starting to get uncomfortable.

Whether AI is “good” or “bad” avoids the massive question underneath it: what happens to the economy when it no longer depends on us, people, in the same way?

Last week, Tristan Harris was on Bill Maher’s show and got closer to this question than most coverage out there.

Maher asked what most of us assume. Isn’t this just about money? Subs, ads, the same old incentives we saw during the advent of the internet and social media.

But Harris pushed back and his answer is worth sitting with:

“If you get all the ChatGPT subscriptions, that doesn’t add up to paying off the amount of debt that they’ve taken. If you get all advertising in the world, that wouldn’t pay off the amount of debt that they’ve taken. The only thing that justifies the amount of investment… is to build artificial general intelligence… to be able to replace every human worker in the economy.”

AI is going to be (and already is) a huge business. Subs, ads, enterprise… all of this is real.

What Harris is getting at is that even this gargantuan business model might not be enough. The scale of investment is so enormous and the returns investors are expecting don’t line up with the productivity gains and improved software tools that are on the horizon.

And once we take that seriously, the implication is straightforward. They’re not just building technology to help people work. They’re building systems that reduce how many people are needed in the first place.

This isn’t a conspiracy. This is where the incentives lead.

There’s an old concept in economics called the resource curse. You often see it in countries rich in oil or minerals. The wealth is real, oftentimes massive, but it changes how their economies work. Instead of depending on people to generate value, the system depends on extraction. Governments invest in infrastructure and production, not in their people. Incentives always win.

Over time, the relationship between the economy and people moves. Slowly at first, then quickly.

AI creates a version of this dynamic for advanced and mature economies. You could call it an intelligence curse. When more of your economic output comes from machines vs. people, the incentives move in that direction. It moves towards energy, infrastructure and compute. And moves away from labor.

What does this look like in practice? It won’t look like a collapse or a sudden wave of unemployment that we can point to.

It looks like companies getting more efficient, producing more with fewer people. Margins improving and growth continuing to go up.

At the same time, headcount goes down. Entire categories of work get compressed. Fewer new roles open up to replace the old ones.

Corporate tax revenues can hold up in this environment. But income tax is another story. Eventually, that will start to soften and the effects will show up locally.

There will be less municipal spending, property values go down. It’ll be small shifts at first, followed by more noticeable ones.

It’s a strange kind of divergence. The economy looks healthy from the top down but feels different on the ground. The economy works, but fewer people do.

This part is easy to miss. This isn’t just about being unemployed. It’s about losing your place in society and in how the economy functions.

Subscribe now

Where might this show up first?

Finance, insurance, law, media, consulting. These kinds of roles have been seen as stable for a long time. High-skill, high-wage jobs that sit at the center of how the modern economy runs.

And people in these roles are supporting local systems all around them: housing markets, public schools, small businesses, local tax revenues.

The parts of the economy that we think of as the strongest right now might actually be the most exposed.

If you look at how policy is responding, it’s mostly focused on the surface. Privacy rules, deepfakes, guardrails around how AI is used. All of this matters, but it just doesn’t touch the core issue.

The real shift is happening underneath. It’s about how output relates to labor. How growth relates to real people. I’ve yet to see a plan for this. Not on the state level or federally.

There’s no playbook we can cite for when a tax base built on human income starts to fade away from the economy generating that income.

Currently, we’re mostly reacting to what AI does and not preparing for what it changes.

Maybe things will move slower than it feels. Maybe there are limits that we’re not seeing yet that keep more people in the system. I don’t know.

But the direction isn’t subtle. This isn’t a dramatic “machines take over” story… it’s much quieter than that. This is about incentives shifting, bit by bit, away from real people.

Once this shift sets in, it’ll be nearly impossible to unwind it.

The economy probably won’t break. It’ll just stop needing all of us.

Leave a comment